Cancer prediction with liquid biopsy using cell-free DNA and immune repertoire

Abstract

Liquid biopsy is a noninvasive method for cancer diagnosis with great academic and practical potential. The immune repertoire and cell-free DNA (cfDNA) in blood is known to reflect the cancer patients’ disease status. In this thesis, two machine learning models were developed from immune repertoire pattern and cfDNA mutation pattern, respectively. Utilizing novel methods such as inclusion of B cell receptor in immune repertoire model and whole genome sequencing (WGS) variation acquisition, both models were able to obtain superior cancer prediction performance. This work paved the way for accurate liquid biopsy cancer detection in pan-cancer manner, even in early cancer.